Training public employees to use ChatGPT is not enough: building real institutional AI capability
· BEXTIO
Learning to write better prompts can be useful.
But an institution does not become capable in artificial intelligence simply because several employees know how to use a conversational assistant.
Institutional capability is broader.
It combines human skills, the quality of data and documents, acceptable-use rules, accountability, security, authorization processes, value measurement and vendor management.
Training that focuses only on the tool can create fragmented use.
Some employees may use personal accounts, others different services; sensitive documents may be copied into uncontrolled environments; and outputs may sometimes be used without sufficient verification.
Training therefore needs to be role-based.
Senior leaders need to understand AI capabilities and limitations, prioritize opportunities, ask the right questions about data, costs, risks and vendors, and decide when a project deserves a pilot.
Managers need to know how to turn a business problem into a use case, define indicators, organize human validation and monitor adoption.
Employees need to learn how to use AI for authorized tasks, verify outputs, protect information, recognize limitations and report errors or incidents.
Technical, security, compliance, procurement and control teams need specialized capability to assess architecture, access, contracts, models, changes and reversibility.
A capable institution also has mechanisms.
It knows which systems are being used.
It assigns owners.
It distinguishes low-risk use from sensitive use.
It records decisions.
It can suspend a system when necessary.
It can change vendors without losing all of its knowledge.
Training then becomes one element of a wider operating system.
BEXTIO recommends building capability progressively: raise awareness, diagnose the current situation, select a small number of use cases, train by role, install the necessary registers and rules, test them through a pilot and then measure the maturity achieved.
The final question is not: “How many people attended training?” It is: “Has the institution become able to use AI in a useful, responsible, traceable and sustainable way?”